A copula-based mixed-effects model for spatio-temporal dependence in fNIRS data
27th International Conference on COMPUTATIONAL STATISTICS (COMPSTAT 2026), Athens, Yunanistan, 25 - 28 Ağustos 2026, ss.1, (Özet Bildiri)
- Yayın Türü: Bildiri / Özet Bildiri
- Basıldığı Şehir: Athens
- Basıldığı Ülke: Yunanistan
- Sayfa Sayıları: ss.1
- Orta Doğu Teknik Üniversitesi Adresli: Evet
Özet
Functional near-infrared spectroscopy (fNIRS) records oxyhemoglobin changes across multiple scalp channels simultaneously, generating data with three dependence structures that standard linear mixed-effects models cannot capture. Spatially, adjacent channels share overlapping cortical territories and common vascular supply, so hemodynamic responses co-vary with inter-sensor distance. Temporally, the hemodynamic response function duration exceeds typical inter-trial intervals, causing carry-over between consecutive trials. Marginally, residuals are non-Gaussian due to the asymmetric onset-to-recovery shape of the hemodynamic response and heavy-tailed physiological noise from cardiac and respiratory processes. Ignoring these structures inflates Type I error rates, biases fixed-effect estimates, and degrades predictive accuracy. A copula-based framework decouples these three structures via Sklar's theorem: skew-t marginals accommodate asymmetry and tail heaviness per channel independently of dependence, a Gaussian copula with a Matern-3/2 kernel captures spatially-decaying inter-channel correlation through a single range parameter $\phi$, and Archimedean copulas model asymmetric temporal carry-over with lower-tail dependence. The framework is applied to a 20-channel LIGHTNIRS dataset and compared against mixed-effects benchmarks, demonstrating that modelling spatio-temporal dependence is the primary driver of predictive improvement in multi-channel fNIRS data.